On Sun, Nov 29, 2020 at 06:34:01PM +0100, Paolo Bonzini wrote:
On 29/11/20 05:13, Sasha Levin wrote:
Which doesn't seem to be suitable for stable either... Patch 3/5 in
Why not? It was sent as a fix to Linus.
Dunno, 120 lines of new code? Even if it's okay for an rc, I don't see why it is would be backported to stable releases and release it without any kind of testing. Maybe for 5.9 the chances of breaking
Lines of code is not everything. If you think that this needs additional testing then that's fine and we can drop it, but not picking up a fix just because it's 120 lines is not something we'd do.
things are low, but stuff like locking rules might have changed since older releases like 5.4 or 4.19. The autoselection bot does not know that, it basically crosses fingers that these larger-scale changes cause the patches not to apply or compile anymore.
Plus all the testing we have for the stable trees, yes. It goes beyond just compiling at this point.
Your very own co-workers (https://cki-project.org/) are pushing hard on this effort around stable kernel testing, and statements like these aren't helping anyone.
If on the other hand, you'd like to see specific KVM/virtio/etc tests as part of the stable release process, we should all work together to make sure they're included in the current test suite.
Maybe it's just me, but the whole "autoselect stable patches" and release them is very suspicious. You are basically crossing fingers
Historically autoselected patches were later fixed/reverted at a lower ratio than patches tagged with a stable tag. I *think* that it's because they get a longer review cycle than some of the stable tagged patches.
and are ready to release any kind of untested crap, because you do not trust maintainers of marking stable patches right. Only then, when a
It's not that I don't trust - some folks forget, or not realize that something should go in stable. We're all humans. This is to complement the work done by maintainers, not replace it.
backport is broken, it's maintainers who get the blame and have to fix it.
What blame? Who's blaming who?
Personally I don't care because I have asked you to opt KVM out of autoselection, but this is the opposite of what Greg brags about when he touts the virtues of the upstream stable process over vendor kernels.
What, that we try and include all fixes rather than the ones I'm paid to pick up?
If you have a vendor you pay $$$ to, then yes - you're probably better off with a vendor kernel. This is actually in line (I think) with Greg's views on this (http://kroah.com/log/blog/2018/08/24/what-stable-kernel-should-i-use/).